Re: [PATCH v4] Add support for the readnever concept

[Sergio Durigan Junior <sergiodj@redhat.com>]

On Thursday, November 30 2017, Pedro Alves wrote:

> On 11/30/2017 12:25 AM, Sergio Durigan Junior wrote:
>
>> - Split off the fix to 'symbol-file' command argument parsing.  This
>>   patch depends on it
>>   (<https://sourceware.org/ml/gdb-patches/2017-11/msg00799.html>).
>
> This info above conflicts with ...
>
>> While implementing the code for the 'symbol-file' command, I noticed a
>> bug in 'symbol_file_command': GDB adds the symbol file before
>> finishing parsing all the options, which means that the position of an
>> option in the command impacts whether it will be considered or not.  I
>> changed the code there in order to only add the symbol file after all
>> options have been parsed.
>
> ... this info here.  The patch still has that bit...

You're right, I'll remove this bit since there's a whole new patch (with
all its glorious versions) to implement that.

>> 
>> It's also worth mentioning that this patch tests whether GDB correctly
>> fails to initialize if both '--readnow' and '--readnever' options are
>> passed.  We didn't have any infrastructure to test that, so I added a
>> new version of 'gdb_spawn', called 'gdb_spawn_ignore_error', which can
>> be used to start a GDB that will fail without compromising the entire
>> testcase.
>
> This is stale.

Removed the part that refers to 'gdb_spawn_ignore_error'.

>> gdb/testsuite/ChangeLog:
>> 
>> 2017-11-29  Joel Brobecker  <brobecker@adacore.com>
>> 	    Sergio Durigan Junior  <sergiodj@redhat.com>
>> 	    Pedro Alves  <palves@redhat.com>
>> 
>> 	* gdb.base/readnever.c, gdb.base/readnever.exp: New files.
>> 	* lib/gdb.exp (default_gdb_spawn): Add 'ignore_error'
>> 	parameter.  Handle case when 'ignore_error' is set.
>> 	(gdb_spawn_ignore_error): New function.
>
> Also stale...

Removed the part about lib/gdb.exp.

>> @@ -699,7 +700,8 @@ coff_symfile_read (struct objfile *objfile, symfile_add_flags symfile_flags)
>>  	}
>>      }
>>  
>> -  bfd_map_over_sections (abfd, coff_locate_sections, (void *) info);
>> +  if (!readnever_symbol_files && !(objfile->flags & OBJF_READNEVER))
>> +    bfd_map_over_sections (abfd, coff_locate_sections, (void *) info);
>
> I realized something: is it really ever possible to get here with
>
>  readnever_symbol_files && !(objfile->flags & OBJF_READNEVER)
>
> ??
>
> If readnever_symbol_files is set, then it'll have been propagated to
> objfile->flags before we get here, no?

Hm, I think so, yes.  I mean, assuming that objfile != NULL when we call
the readers, then yes; we always call symbol_file_add_with_addrs which
is responsible for propagating the flag.

> So references to the global readnever_symbol_files can (and should)
> be removed from the readers, right?  This removes the need for
> the new '#include "top.h"'s too, so please remove those from
> the patch along the way.

OK, done.

>>  
>>    if (info->stabsects)
>>      {
>> diff --git a/gdb/doc/gdb.texinfo b/gdb/doc/gdb.texinfo
>> index 4f552225aa..a3127dd637 100644
>> --- a/gdb/doc/gdb.texinfo
>> +++ b/gdb/doc/gdb.texinfo
>> @@ -1037,6 +1037,15 @@ Read each symbol file's entire symbol table immediately, rather than
>>  the default, which is to read it incrementally as it is needed.
>>  This makes startup slower, but makes future operations faster.
>>  
>> +@item --readnever
>> +@cindex @code{--readnever}, command-line option
>> +Do not read each symbol file's symbolic debug information.  This makes
>> +startup faster but at the expense of not being able to perform
>> +symbolic debugging.  DWARF unwind information is also not read,
>> +meaning backtraces may become incomplete or inaccurate.  One use of
>> +this is when a user simply wants to do the following sequence: attach,
>> +dump core, detach.  Loading the debugging information in this case is
>> +an unnecessary cause of delay.
>>  @end table
>>  
>>  @node Mode Options
>> @@ -18576,6 +18585,14 @@ tables by using the @samp{-readnow} option with any of the commands that
>>  load symbol table information, if you want to be sure @value{GDBN} has the
>>  entire symbol table available.
>>  
>> +@cindex @code{-readnever}, option for symbol-file command
>> +@cindex never read symbols
>> +@cindex symbols, never read
>> +@item symbol-file @r{[} -readnever @r{]} @var{filename}
>> +@itemx file @r{[} -readnever @r{]} @var{filename}
>> +You can instruct @value{GDBN} to never read the symbolic information
>> +contained in @var{filename} by using the @samp{-readnever} option.
>
> Maybe copy the rationale for this found in "--readnever" here
> or xref the "--readnever" command line option?

I used an xref.

>
>>  #define READNOW_HELP \
>>    "The '-readnow' option will cause GDB to read the entire symbol file\n\
>>  immediately.  This makes the command slower, but may make future operations\n\
>> -faster."
>> +faster.\n"
>> +
>> +#define READNEVER_HELP \
>> +  "The '-readnever' option will prevent GDB from reading the symbol file's\n\
>> +symbolic debug information."
>>  
>
> Why not merge these two macros in one?  We'll always want to
> use them together, in the same order?  You could rename the
> macro to READNOW_READNEVER_HELP for example.

Did it.

I'll wait until we sort out the patch to extend the argument parser of
'{add-,}symbol-file' before I submit v5.

Thanks,

-- 
Sergio
GPG key ID: 237A 54B1 0287 28BF 00EF  31F4 D0EB 7628 65FC 5E36
Please send encrypted e-mail if possible
http://sergiodj.net/